Cadences and Follow-Up: What the Evidence Says

What the data says, and how much to trust it

Almost all follow-up data comes from vendors who sell outreach software, measured on their own customers. That does not make it useless, but it does mean you should read it as a description of those customers' campaigns, not as a law of nature.

The most cited recent dataset shows why. The agency Belkins analyzed 7.5 million sent emails across 2025 for its 2026 study. The first email had the highest reply rate of any single step, about 0.59%, but follow-ups together produced more replies than the opener: steps two to six accounted for 58.6% of all replies. Treat the exact percentages with care: figures like these move with the sample, the year and the mailbox providers' filtering, and the published benchmarks disagree with each other about how many follow-ups are worth sending.

Instantly, which sells cold-email and AI sales-agent tools, reported a platform-wide average reply rate of 3.43% for 2025 in its 2026 benchmark, with the top 10% of senders above 10.7%. It is explicit that these are platform averages, not controlled experiments. Reply rates in the low single digits are the backdrop against which every cadence decision is made.

Two conclusions survive the uncertainty. Follow-up matters: a cadence of one email leaves most of the available replies on the table. And each additional step earns less than the one before, while the cost of being unwelcome does not fall.

The asymmetry: replies versus complaints

Every follow-up has a small chance of a reply and a small chance of something worse: a spam complaint, an unsubscribe, or a buyer who quietly decides never to deal with you. Those costs fall on the sending domain, not just the campaign.

The mailbox providers have made that cost explicit. Google's sender guidelines require bulk senders (5,000 or more messages a day to personal Gmail accounts) to keep the spam rate reported in Postmaster Tools below 0.30%, and recommend staying under 0.10%. Yahoo's requirements, enforced from February 2024, include keeping the spam rate below 0.3% and honouring unsubscribes within two days. Exceed those limits and messages start going to spam or being rejected, which damages every email the domain sends, including replies to existing customers.

So a cadence design is a trade: each extra step must earn enough replies to justify its added complaint risk. A reasonable starting point many teams use is three to five steps over two to three weeks, each adding something new (a different angle, a useful resource, a direct question) rather than 'just bumping this'. Kavanah does not prescribe a number; measure your own.

When to stop: three rules

Stop when they reply. A reply of any kind ends the automated sequence; the conversation is now a human one. Continuing to send the scheduled step after someone has answered is the most damaging cadence error there is.

Stop when you meet. A booked meeting is a reply in another channel.

Stop when they ask you to. An unsubscribe, a 'please remove me', a complaint. In the US, the FTC's CAN-SPAM guidance requires opt-out requests to be honoured within 10 business days, and Yahoo asks bulk senders to honour them within two days. Lesson 3.4 covers the rules in full.

Kavanah builds all three into its cadences. A reply attributed to an enrolled contact stops the enrollment and cancels any step already queued but not yet sent. A personal calendar meeting with the contact (a timed event with at most 12 people) stops enrollments that began before the meeting. A reply that reads as an opt-out request, a deliberately conservative match on the new text of the reply, adds the address to the workspace's suppression list and cancels every active enrollment to that address. And the suppression list is checked twice: before a step is generated and again at the moment of sending, because the opt-out that matters is often the one that arrives after a step was scheduled.

Building and testing a cadence in Kavanah

A Kavanah cadence is an ordered list of one to eight email steps, each a number of days after the previous one; the first step's delay is measured from enrollment. Subjects and bodies can use placeholders such as the contact's first name, the deal name, the sender's name and a booking link. Creating a cadence sends nothing. Sends start only when someone enrolls a contact, and enrolling is classed as an outbound send: it is removed from the agent in safe mode, and in full mode it follows your action policy for sending messages, which by default parks it for approval (Lesson 3.4).

A cadence can be a deal cadence, attached to a contact on a deal, or a network cadence for outreach before any deal exists; network cadences always carry an unsubscribe link and cannot turn it off. Cadence steps are generated by a runner every few minutes and delivered through a member's connected mailbox (for a deal cadence, the enrolling member's own), within the per-mailbox daily caps set in Settings, Outbound. For network cadences, Kavanah also warns when a campaign is about to run out of people to contact, because a cadence that runs dry keeps sending follow-ups while starting no new conversations, and nothing looks broken.

To test, give a step up to four subject-line variants. Each enrolled contact is assigned one arm at enrollment and keeps it for the whole sequence, so a reply can be attributed. Read results with get_cadence_variant_stats, which reports sends, replies and reply rate per arm. Compare rates, never raw counts, because arms rarely receive equal volume, and do not call a winner until each arm has enough sends that one or two replies would not flip the result.